Lore

Inspiration

Every team has a wiki. And every team knows their wiki is probably wrong. Not because people are careless - but because the update already happened. A decision changed in Slack. Someone figured out the real process and shared it in a thread. A policy shifted after a meeting. The knowledge exists, it just never made it back to the doc.

I wanted to fix that without asking people to change how they work.

What it does

Lore is a Slack agent that cross-checks your wiki against your Slack history in real time. Ask it a question and it searches both sources simultaneously - your Notion, Confluence, or Google Drive, and your Slack threads. If they agree, it answers and cites both. If they contradict each other, it tells you exactly where they diverge and drafts the doc update for you. One click to push it back to the wiki.

You can also trigger it directly from a thread - reply with @lore update this and it reads the conversation, figures out what changed, and drafts the edit.

How we built it

Lore is built on three technologies:

• Real-Time Search API: searches Slack thread history by topic to surface relevant conversations
• MCP: connects to Notion, Confluence, and Google Drive for both reading and writing back
• Claude: compares both sources, detects contradictions, drafts minimal before/after edits, and ranks experts with cited evidence

The Slack layer is built with Bolt for TypeScript in Socket Mode. Every response is a Block Kit card, color coded by outcome so you know at a glance what Lore found.

No data is stored. Everything happens live at query time.

Challenges we ran into

Getting the conflict detection right took the most iteration. The prompt needs to distinguish between sources that genuinely contradict each other versus ones that are just talking about different aspects of the same thing.

Accomplishments that we're proud of

The direct update flow: replying to a thread with @lore update this and having it read the conversation, find the relevant doc, and draft a minimal edit that matches the existing formatting.

What we learned

The hardest part was the reconciliation. Searching Slack and searching a wiki are both solved problems. Getting a model to reason about whether two sources on the same topic agree or disagree, and then draft a minimal fix that respects the existing doc's voice, is where most of the work went.

Also: human-in-the-loop design matters a lot here. Lore never updates a doc without showing you the before/after first. That one decision makes it feel trustworthy rather than scary.

What's next for Lore

Right now Lore connects to one wiki at a time. Supporting multiple sources simultaneously is the obvious next step.

Proactive conflict detection - Lore noticing drift in the background and surfacing it without being asked - is the bigger vision. Right now it responds when you ask. The goal is for it to tell you before you need to ask.

Built With

• bolt
• mcp
• notion
• rts
• slack